Skip to content Skip to main navigation Report an accessibility issue

Machine and Deep Learning for materials: accelerating imaging, deriving physics, and online control for synthesis

Dr. Rama Vasudevan
R&D Associate & Data Analytics Coordinator
Center for Nanophase Materials Sciences 
Friday, November 1, 2019   2:30-3:30pm
John Tickle Building Room 410


Machine learning (ML) approaches have attracted attention in the recent past due in large part to successes of deep learning methods in traditionally challenging domains such as natural language processing, computer vision and interactive gameplay. However, application of ML in traditional physical sciences is still in its infancy. One reason is that the correlative nature of ML methods lies in stark contrast to rigorous physics-derived models that are the mainstay of most scientists in physics and related fields.


In this talk, I will discuss our use of statistical and machine learning to various forms of microscopy and materials synthesis conducted at the Center for Nanophase Materials Sciences. These methods can accelerate imaging acquisition, improve resolution, derive correlations in high dimensional data, and find optimal synthesis routes. Collectively, these can accelerate knowledge generation and aid in physics extraction, and ideally, reduce the time from computational discovery to realized materials. Towards this aim, I will touch on some recent work within our group for autonomous synthesis and control of scientific instrumentation to rapidly yield actionable insight into a variety of physical systems.


Rama Vasudevan is an R&D Associate and Data Analytics Coordinator at the Center for Nanophase Materials Sciences and works within the Scanning Probe Microscopy group. He completed his PhD in Materials Science at the University of New South Wales, graduating in early 2013 before moving to the CNMS as a postdoctoral researcher. In 2016, he was hired as a staff member. His interests lie in the intersection of machine learning and microscopy, for deriving physics from multidimensional datasets. His primary materials science interests are ferroelectrics and complex oxides, as well as developing techniques for scanning probe microscopy to better investigate these materials at the nanoscale.